×

You can configure live migration settings to ensure that the migration processes do not overwhelm the cluster.

You can configure live migration policies to apply different migration configurations to groups of virtual machines (VMs).

Configuring live migration limits and timeouts

Configure live migration limits and timeouts for the cluster by updating the HyperConverged custom resource (CR), which is located in the kubevirt-hyperconverged namespace.

Prerequisites
  • You have installed the OpenShift CLI (oc).

Procedure
  • Edit the HyperConverged CR and add the necessary live migration parameters:

    $ oc edit hyperconverged kubevirt-hyperconverged -n kubevirt-hyperconverged

    Example configuration file:

    apiVersion: hco.kubevirt.io/v1beta1
    kind: HyperConverged
    metadata:
      name: kubevirt-hyperconverged
      namespace: kubevirt-hyperconverged
    spec:
      liveMigrationConfig:
        bandwidthPerMigration: 64Mi
        completionTimeoutPerGiB: 800
        parallelMigrationsPerCluster: 5
        parallelOutboundMigrationsPerNode: 2
        progressTimeout: 150
        allowPostCopy: false

    where:

    bandwidthPerMigration

    Specifies the bandwidth of each migration in bytes per second. For example, a value of 2048Mi means 2048 MiB/s. Default: 0, which is unlimited.

    completionTimeoutPerGiB

    Specifies the length of time, in seconds per GiB of memory, at which the migration is canceled if it has not completed. For example, a VM with 6GiB memory times out if it has not completed migration in 4800 seconds. If the Migration Method is BlockMigration, the size of the migrating disks is included in the calculation.

    parallelMigrationsPerCluster

    Specifies the number of migrations running in parallel in the cluster. Default: 5.

    parallelOutboundMigrationsPerNode

    Specifies the maximum number of outbound migrations per node. Default: 2.

    progressTimeout

    Specifies the length of time, in seconds, at which the migration is canceled if memory copy fails to make progress. Default: 150.

    allowPostCopy

    Specifies whether the post copy mode is enabled. You can enable post copy mode to allow the migration of one node to another to converge, even if a VM is running a heavy workload and the memory dirty rate is too high. By default, allowPostCopy is set to false.

    You can restore the default value for any spec.liveMigrationConfig field by deleting that key/value pair and saving the file. For example, delete progressTimeout: <value> to restore the default progressTimeout: 150.

Configure live migration for heavy workloads

When migrating a VM running a heavy workload (for example, database processing) with higher memory dirty rates, you need a higher bandwidth to complete the migration.

If the dirty rate is too high, the migration from one node to another does not converge. To prevent this, enable post copy mode.

Post copy mode triggers if the initial pre-copy phase does not complete within the defined timeout. During post copy, the VM CPUs pause on the source host while transferring the minimum required memory pages. Then the VM CPUs activate on the destination host, and the remaining memory pages transfer into the destination node at runtime.

Configure live migration for heavy workloads by updating the HyperConverged custom resource (CR), which is located in the kubevirt-hyperconverged namespace.

Prerequisites
  • You have installed the OpenShift CLI (oc).

Procedure
  1. Edit the HyperConverged CR and add the necessary parameters for migrating heavy workloads:

    $ oc edit hyperconverged kubevirt-hyperconverged -n kubevirt-hyperconverged

    Example configuration file:

    apiVersion: hco.kubevirt.io/v1beta1
    kind: HyperConverged
    metadata:
      name: kubevirt-hyperconverged
      namespace: kubevirt-hyperconverged
    spec:
      liveMigrationConfig:
        bandwidthPerMigration: 0Mi
        completionTimeoutPerGiB: 150
        parallelMigrationsPerCluster: 5
        parallelOutboundMigrationsPerNode: 1
        progressTimeout: 150
        allowPostCopy: true

    where:

    bandwidthPerMigration

    Specifies the bandwidth of each migration in bytes per second. The default is 0, which is unlimited.

    completionTimeoutPerGiB

    Specifies the length of time, in seconds per GiB of memory, at which the migration is canceled if it has not completed and post copy mode is triggered, if enabled. You can lower completionTimeoutPerGiB to trigger post copy mode earlier in the migration process, or raise the completionTimeoutPerGiB to trigger post copy mode later in the migration process.

    parallelMigrationsPerCluster

    Specifies the number of migrations running in parallel in the cluster. The default is 5. Keeping the parallelMigrationsPerCluster setting low is better when migrating heavy workloads.

    parallelOutboundMigrationsPerNode

    Specifies the maximum number of outbound migrations per node. Configure a single VM per node for heavy workloads.

    progressTimeout

    Specifies the length of time, in seconds, at which the migration is canceled if memory copy fails to make progress. Increase this parameter for large memory sizes running heavy workloads.

    allowPostCopy

    Specifies whether the post copy mode is enabled. You can enable post copy mode to allow the migration of one node to another to converge, even if a VM is running a heavy workload and the memory dirty rate is too high. Set allowPostCopy to true to enable post copy mode.

  2. Optional: If your main network is too busy for the migration, configure a secondary, dedicated migration network.

    Post copy mode can impact performance during the transfer, and should not be used for critical data, or with unstable networks.

Live migration policies

You can create live migration policies to apply different migration configurations to groups of VMs that are defined by VM or project labels.

You can create live migration policies by using the OKD web console.

Creating a live migration policy by using the CLI

You can create a live migration policy by using the command line.

KubeVirt applies the live migration policy to selected virtual machines (VMs) by using any combination of labels:

  • VM labels such as size, os, or gpu

  • Project labels such as priority, bandwidth, or hpc-workload

For the policy to apply to a specific group of VMs, all labels on the group of VMs must match the labels of the policy.

If multiple live migration policies apply to a VM, the policy with the greatest number of matching labels takes precedence.

If multiple policies meet this criteria, the policies are sorted by alphabetical order of the matching label keys, and the first one in that order takes precedence.

Prerequisites
  • You have installed the OpenShift CLI (oc).

Procedure
  1. Edit the VM object to which you want to apply a live migration policy, and add the corresponding VM labels.

    1. Open the YAML configuration of the resource:

      $ oc edit vm <vm_name>
    2. Adjust the required label values in the .spec.template.metadata.labels section of the configuration. For example, to mark the VM as a production VM for the purposes of migration policies, add the kubevirt.io/environment: production line:

      apiVersion: migrations.kubevirt.io/v1alpha1
      kind: VirtualMachine
      metadata:
        name: <vm_name>
        namespace: default
        labels:
          app: my-app
          environment: production
      spec:
        template:
          metadata:
            labels:
              kubevirt.io/domain: <vm_name>
              kubevirt.io/size: large
              kubevirt.io/environment: production
      # ...
    3. Save and exit the configuration.

  2. Configure a MigrationPolicy object with the corresponding labels. The following example configures a policy that applies to all VMs that are labeled as production:

    apiVersion: migrations.kubevirt.io/v1alpha1
    kind: MigrationPolicy
    metadata:
      name: <migration_policy>
    spec:
      selectors:
        namespaceSelector:
          hpc-workloads: "True"
          xyz-workloads-type: ""
        virtualMachineInstanceSelector:
          kubevirt.io/environment: "production"

    where:

    namespaceSelector

    Specifies the project labels.

    virtualMachineInstanceSelector

    Specifies the VM labels.

  3. Create the migration policy by running the following command:

    $ oc create -f <migration_policy>.yaml